Академический Документы
Профессиональный Документы
Культура Документы
‘cos(2pn/q) logl2 - 2 cos(27n/q)].
The prime indicates that in case of an even g only one-half of the last
term shall be taken in the sum. Thus for a positive proper fraction z the
value of y(z) can be expressed as a finite combination of elementary
functions. By means of (10) this result may be extended to every rational
value of z. This is Gauss’ theorem,
1.7.4, Some infinite series connected with the y function
If we define
Af) =fle+D=f, A*f(2)= § (A)ieena ens
it follows from 1,7(8) that
Avla+z)=V(a+2),
so that we have
A? datz) = AL1/(a +2)] =~ 1/[(a +2) (a+24+)]
and
A™y(a +2) =A™' [V/(a+z)]
=D" (n-D!/[(atz) (a+z4+ Dos (o+z4+n-D
Hence the development of ys(a +2) in a factorial series is convergent for
Re(a + 2) > 0, @ not a negative integer, and is of the form (Nérlund 1924,
p- 261)
2 1 az-2) 1 ez-D(z-2)
us =¢ ——— frog ied Eeuec nL
OO ee seen s a ee)
The functional equation 1.7(10) is useful for summing some series.
We have for instance:
(31) 2 (a +mb)7' =b7! 3 (m +a/b)7' = 67" [hb (n+ 14+0/b) -y (a/d)],.
2 =
1 1 ue :$ ‘a—b “Wa oe
eh ae oh com eas 2 ce) lon
a 1 a a a
wuvsfo(gpened)-o(Se3-o(Gened) Ge}2 SPECIAL FUNCTIONS 17.4
and, ifn +e,
1 1 1
SS
a+b a42b a+3b
mee = BOWL ab mr (444 aby,
18. The function G(z)
‘The function G(z) is defined by
(CD) G(z) = Uh + ez) - WChz)
From 1.7(13)and L7(14) we have
(Q G@)=2f 4d de Rez>0,
(3) G)=2 fe (sey de Rez>0,
A consideration of { e~##(1+ 7 *)~' de extended over the contour used
in deriving 1.5(31) yields the more general representation
Clare ease arta
~ Yin 0.
Since (Whittaker-Watson, 1927, p. 249)
(3) le mth + (et 17]! eo tde = 1—% log(2z),
we have Binet’s first expression of log I'(z),
(4) log P(z) =(z —%) log z —z +% log(27)
+ fPllet- Yt a yet e Bade Rez>0,
or, more generally [ cf, 1.5(1) and also 1,7(25), 1.7(26)],
(5) logI"(z) =(z — 8) log z~z +% log(2n)
+ Jee ett art ay et en Bde
~kn0.
From 1.7(8) we have
a log Mg
dz?
(10) & (2) = = Viz+2)%
azo
Now we make use of a summation formula due to Plana (Lindeléf, 1906,
p61),
Cay Ef) = (C0) + J flr) devi LC pled ~ fa" DT a,
valid if
Ll) f(2) is regular for Re £ > 0, C=rtit,
2 lim eo ldl f(r 448) = 0 uniformly for 0. 1 when z is on
the real axis to the right of | or to the left of ~ 1.
From 3.2 (32) we have for the same function
(9) e7 #" QH @ 19)
_ PC +y +p) 0p) 4” be
2 2rd +v-p) z+1
x F[-v, l+vjltus 4-2
Kw
+%0P@) (3) Flav, Lt v3 1-05 (1 ~ ale
=
From (19), 3.2(32), and 3.2(3)
(20) Q# (2, 1-) - e7 #7 Q4 (2) = rie #7 PY (2).
Hence, by means of (18) and 3,2 (44)
Tw + 3/2)
(ge yh — EE pe
(2 @n)¥ (F -Ik EE PEO
z+(t~ »"|
2(z2— 1)4
24%
aie" Ee (at — WHO FL ay, ps 9/2 4
: 2(z? - 1)*
ale + (2? ~ 41% af By Wm w5 8/2 +53.47 LEGENDRE FUNCTIONS 143
Furthermore from (9), (20), and 3.2 (44)
(22) Q# (2) sin[n(v +1 -
= (4n)% Pip =v) (22 - )¥ & #7 9) cos(um) [z + (z? = D4]
(2> 0% ;
| dem aom iors ESTE] conta 27
a ~z+(z?@-1)*
* Len(a? 14] ef +s Bons hos =e |}
The expressions (21) and (22) are not of the type 3,2(9) and 3.2(10).
3.4: | Legendre functions on the cut
In many applications of Legendre functions z = x, where -1 (@),, - =niB bee
min
(OA).
H Glan =n Pla gm yn
SS eae -) (3), mint
(
(38) Hla 3% y= Clencn yt yt
©, mint
5.7.1
Iel 0,
x+y <1 with the weight function x71 y¥ ~!,
The series
@) x" y"FiC¥m, -4n, 4-4, kn, —m-n—hs +h; 27%, ¥7*)
(which can be expressed also in terms of F, ) and
(5) x" y" Fyl-Km, Yn, 4m, 4—-n, hs +h; @? +y?-D 27%,
G+y?-Dy?]
occur in the investigation of hyperspherical harmonics.
All these special cases are investigated in the monograph by Appell
and Kampé de Fériet.
5.14. Further series
Hypergeometric series of higher order than two, in two variables,
have been studied by Mellin, Birkeland, Kampé de Fériet (cf. Appell
and Kampé de Fériet 1926 Chapter IX), and by Burchnall and Chaundy
(1941), Burehnall (1942), and Chaundy (1942), Hypergeometric series
in three variables have been investigated by Horn (1889); series in n
variables by Lauricella (Appell and Kampe’ de Fériet Chapter VII) and
by Erdélyi (1937, 1939a). Particular cases of Lauricella’s series occur
in the investigation of hyperspherical harmonics.
An extension of basic hypergeometric series to two variables has
been given by Jackson (1942, 1944).246 SPECIAL FUNCTIONS
REFERENCES
Appell, Paul and M, J. Kampé-de Fériet, 1926: Fonctions hypergeémétriques
et hypersphériques. Polynomes d’ Hermite. Gauthier-Villars.
Bailey, W. N., 1935+ Generalized hypergeometric series (Cambridge Tracts in
Mathematics and Mathematical Physics No. 32) Cambridge,
Barnes, E. W., 1907: Proc. London Math, Soc. (2) 5, 59-116.
Barnes, E. W., 1908: Proc. London Math, Soc. (2) 6, 141-177.
Bames, E. W., 1908: Trans, Cambridge Philos. Soc. 20, 253-279.
Birkeland, Richard, 1927: C. R. Acad. Sci. Paris 185, 923-925,
Borngiisser, Ludwig, 1933: Uber hypergeometrische Funktionen zweier Ver-
ahderlichen, Dissertation, Darmstadt.
Burchnail, J. L., 1939: Quart. J. Math, Oxford Ser. 10, 145-150.
Burchnall, J. L., 1942: Quart. J. Math., Oxford Ser. 13, 90-106.
Burchnall, J. L., and T. W. Chaundy, 1940: Quart, J. Math, Oxford Ser. 11,
249-270.
Burchnall, J. L., and T. W. Chaundy, 1941: Quart, f. Math, Oxford Ser. 12,
112-128,
Chaundy, T. W., 1942: Quart. J, Math., Oxford Ser. 13, 159-171,
Erdélyi, Arthur, 1937: Akad. Wiss. Wien, S-B Ila, 146, 431-467,
Erdélyi, Arthur, 1939: Proc. Roy. Soc. Edinburgh, 59, 224-241,
Erdélyi, Arthur, 1939.a: Nieuw Arch, Wiskunde (2) 20, 1-34.
Erdélyi, Arthur, 1940; Proc, Roy, Soc. Edinburgh 60, 344-361.
Erdélyi, Arthur, 1942; Quart. J. Math, Oxford Ser. 11, 68-77.
Erdélyi, Arthur, 1948: Proc. Roy. Soc. Edinburgh Sect. A 62, 378-385.
Horn, Jakob, 1889: Math, Ann. 34, 544-600.
Horn, Jakob, 1931: Math, Ann 105, 381-407.
Horn, Jakob, 1935: Math, Ann, 111, 638-677.
Hom, Jakob, 1936: Math, Anum 213, 242-291,
Hom, Jakob, 1938: Math, Ann. 115, 435-455,
Hom, Jakob, 1938 a: Monatsh, Math. Phys. 47, 186-194,
Hom, Jakob, 1939: Monatsh. Math, Phys. 47, 359-379.
Ince., E, L., 1942: Proc. Roy. Soc. Edinburgh, A 61, 195-209.
Jackson, F, H. 1942: Quart J. Math, 13, 69-82,
Jackson, F. H., 1944: Quart. J. Math, 15, 49-61.
MacRobert, T. M., 1937-38: Proc. Roy. Soc. Edinburgh 58, 1-13,GENERALIZATIONS OF THE HYPERGEOMETRIC FUNCTION 247
REFERENSES
MacRobert, T. M., 1941: Philos, Mag. (7), 31, 254-260.
MacRobert, T. M., 1942: Quart. J. Math, Oxford Ser. 12, 65-68.
MacRobert, T. M., 1948: Philos Mag. (7) 39, 466-471.
Meijer, C. S., 1936: Nieuw Arch, Wiskunde (2) 18, 10-39.
Meijer, C. S., 1940: Nederl, Akad, Wetensch., Proc. 43, 198-210 and 366-378.
Meijer, C. S., 19414: ibidem 44, 82-92, 186-194, 298-307, 442-451 and 590-598,
Meijer, C. S., 1941b: ibidem 44, 435-441 and 599-605.
Meijer, C. S., 1941¢: ibidem 44, 1062-1070,
Meijer, C. S., 1946: ibidem 49, 344-456, 457-469, 632-641, 765-772, 936-943,,
1063-1072, 1165-1175.
Ore, Oystein, 1929: Cr R. Acad, Sci, Paris, 189, 1238-1240.
Ore, Oystein, 1930: J. Math, Pura and Appl. (9) 9, 311-326.CHAPTER VI
CONFLUENT HYPERGEOMETRIC FUNCTIONS
6.1, Orientation
If we put z= x/d in Gauss” hypergeometric series
For jo 2 ee
oe 1-2-cle+
in which we assume that neither a nor ¢ is zero or a negative integer
we obtain a power series in x whose radius of convergence is |b| and
which defines an analytic function with singularities at x = 0, b, and
co. As 6 + o, the limiting case will define an entire function whose
singularity at x =» is a confluence of two singularities of F (a, 6; 0;x/b)
In this manner we are lead to Kummer’s series
In the notation of generalized hypergeometric series 4,1(1) this is
1 F(a; ¢; x), but in this chapter and in the following two chapters (1)
will be denoted by Humbert’s symbol
Bla, o; x).
Sometimes the notation f(a, c, x) is also used.
‘The series (L) satisfies the differential equation
d*y dy
qrte-a fp -an0.
The substitution
(82) yar et®z, a=%—-x+p, cH lt+2Qp
(2) «
reduces (2) to Whittaker’s standard form
dz ee
(4) He 6Cx +Tt =0.
Either of the two equations will be called a confluent hypergeometric
equation, and any solution of either of them a confluent hypergeometric
function; a and c (or « and y) will be called the parameters, x the
variable.
2486.2 CONFLUENT HYPERGEOMETRIC FUNCTIONS 249
hh this chapter, the investigation of confluent hypergeometric func-
tions will be based on (2), but some of the more significant results
will also be given in Whittaker’s notation.
6.2. Differential equations
Equation 6.1(2) is a homogeneous linear differential equation of the
second order whose coefficients are linear functions of the indepen-
dent variable; and it can be shown that every such differential equation
can be integrated in terms of confluent hypergeometric functions. Let
Che d;
WM (@,x+5) = +(jx46) 2+ @r+bdy=0
be the differential equation. Ifa, =a, = a, = 0, the equation has con-
stant coefficients and can be integrated by elementary functions. This
case wil] be excluded. The transformation
Q) yse™z, x=dEtp, (AF 0)
carries (1) into
d*z
@) (06+ BQ) yee +(e £4 By)
dz
a (a, €+ B,) 2=9,
where
(4) ay=ag/A, a,=A'lhh a,=dA(h),
Bo=(a.utb)/M, By= (AA) +B (V/A,
B, = pA(h) + Bh),
AQ) =a,h?+a,h+a,, A'(d)=2a,h+a,,
Blh)=b kh? +b,ht+b,, B'(h)=2bh+d,.
If we can determine A, 1, and h, so that
a,p+b,=0, a,+AA'(A)=0, A(h)=0,
then (3) reduces to 6,1(2), In other cases a change of the independent
variable will reduce (3) to a confluent hypergeometric equation or to
the Bessel equation. The results are set out inthe table on the follow-
ing page, where 5(a, c, x) stands for any solution of 6,1(2), and C, (x)
for any solution of the Bessel equation
ady wy
(5) x on ges (r-v)y=0.
Ix x6.2
SPECIAL FUNCTIONS
250
€ 9%
CV?) == HY an 1D yh Tg] Oe eee |e
yl) 9] =o S ong
(499 D09 PE | 0% to ox |
“Cnay/y) ,F-K =D
(qa)/o~= 4 :
GP1K ME I =. ox'o o=% | (2)
At : (a ®
Cozy) a =?
ce otto :
222 (Cq'o = '9°n) = ' Elon oF 0 ¢*o°op —to= a
ee lee | @
VA a=> ‘ox?
suoyouwseg z ET x y suordunssy
— z 1 one pe z, ‘ °. « ‘
ep qtr vasa 7 Mv So+y'oe va(yy W+RY=¥ ‘ey 2=h
=k Rg 4 x2) + L('q 4 x!o) + ,£(Cq +n) Jo uoronpay6.2 CONFLUENT HYPERGEOMETRIC FUNCTIONS 251
The reduction of (1) to the confluent hypergeometric equation is not
unique, for 6,1(2) can be transformed into an equation of the same form
in several ways. If we put
(10) x=Aé, y=xPeMz,
the differential equation 6,1 (2) tranforms into
d?n dn
age tle# 2p (= 2h) a8)
+[plpte-D/E-Ma~he + p-2hp) +A? Ath 1) El =0,
and this is the differential equation of S(a, y, €) if
(12) p=0 or l-c, &=0 orl, d(1-2k)=1,
=Ma-he)+p, y=e+2p
ql) é
Thus, counting the identity transformation, 6.1(2) can be transformed in
fow different ways into an equation of the same form.
The confluent hypergeometric equation is a homogeneous linear
differential equation of the second order with a singular point of regular
type at x = 0, and with coefficients which are regular for all x 4 0
(includiag x = 00). Every such equation can be shown (Tricomi 1948) to
be of the form
a’y bY dy Boy
2 (ord) 2s (orb ed) g00,
and it is easy to see that [apart from the trivial case when (13) can be
integrated in terms of elementary functions] the integration of (13)
leads to confluent hypergeometric functions or to Bessel functions, We
have to distinguish between two cases of (13).
Ifa? £ 4a, we have
(14) y a278/? 8/2 wlk, py E),
where
(15) x =(B-%ab) (a?- 4a), p= %[(b- 1)? -4y]%,
&=(a?~4a)* x,
and w(x, 2, x) is any solution of 6.1(4); and if a? = 4a, we have
(16) y = 2¥7P/2 6/7 C (8),
(13)
where
(17) v=((6-D?=4y]*, &=20(8 - Kab) x]*.
The equation
d* dy
(is) X2-22 4 (axes BY XX + (0X4 GHP +K)y =0.
xX? ax252 SPECIAL FUNCTIONS 6.2
in which p # 0 is any number, is reduced to (13) by the substitution
(19) =X", a-A/y b=(B+v- DA,
a=Dv?, B=Gv%, yok
and hence can be reduced to the confluent hypergeometric equation. A
more general form can be obtained from (18) by patting
(20) r-¥e0[ feo]
Y satisfies the equation
2
(on x# 22, (axon By 2g x4 [xr scxres
dx? dX
; d¢
eee Bs 24X— =0,
+(AXP+B- Dh 4 G24 “ty 0
For instance, with ¢(X) = 4X7 it follows that
d*z dz
ax? + (AX? + BX7+ Oey
+ (DX + EXP +4 FX™ + GXP4HX°+K) z= 0
can be integrated in terms of confluent hypergeometric functions pro-
vided that
(23) E=%AB, F=\%B?, H=¥%B(C+0-D.
At this point it is worth mentioning that Bessel functions themselves
are special confluent hypergeometric functions (cf. section 6.9.1) so
that in all non-trivial cases the differential equations of this section
can be integrated in terms of confluent hypergeometric functions.
‘The principal linear differential equations of the second order
which can be integrated in terms of confluent hypergeometric functions
are (1), (13), (18), and (22), the last under conditions (23).
(22) x?
6.3. The general solution of the confluent equation wear the origin
We shall investigate the confluent hypergeometric equation in the
form
d® d)
Weed 2 -ay=0.
One solution of this equation is
Q) y,= Og, ¢; x),64 CONFLUENT HYPERGEOMETRIC FUNCTIONS 253
and the transformations 6.2(10), 6.2(12) give three further solutions,
viz,
(3) y, 221° @(a-c 41, 2-c; x),
(4) y,=e* B(e-a c5~x),
6) y,=2'e*O(1-a, 2-0; -2),
From their behavior at the origin it follows that y, and y, are line-
arly independent if ¢ is not an integer so that (in this case) the general
solution of (1) may be written as
(6) y=dy, + By,
where A and B are arbitrary constants. The exceptional case of an in-
teger c will be discussed later (cf. section 6.7.1).
On the other hand, both y, and y, are solutions of (1), both are
regular at the origin, and have the value unity there. Since (with a non=
integer c) the differential equation cannot have more than one such
solution, we must have y, =7, or
(7) ®(a, ¢; x) = e* O(c —a, c; =x).
This is known as Kummer’s eransformation. Similarly y, = y, by the
Kummer transformation,
Kummer’s transformation is a limiting case of Euler’s transformation
F (a, bs ¢3 2) = (1-2) Fle ~ a, 6; ¢; zz - DI
of the hypergeometric series (cf. 2,1(22)], and is a very important
relation. Our proof assumes that ¢ is not an integer, but the trans-
formation holds, by continuity, for positive integer c.
ify, andy, are any two solutions of (1), their Wronskian,
d
ge cae
must be of the form K__ e* x ~*, where the constant K can be evaluat-
ed by using the first two terms of the series expansions of the y’s. We
find
(9) Wg = Way = yg =~ Wyq = (1-0) x8 e%
All the other Wronskians of the four solutions vanish identically,
6.4, Elementary relations for the ® function
As in the theory of Gauss’ hypergeometsic series, the four functions
@ G@)=@+Le;x), Oa-)=O(-1,¢;%),
O(c) =O(a,c+1;x), O(c -)= O(a, c-1;x)254 SPECIAL FUNCTIONS 6.4
are said to be contiguous to ® = ®(a, c; x), The function ® andany two
functions contiguous to it are linearly connected. The six formulas
describing these connections can be derived from Gauss’ relations
between the contiguous functions (cf. section 2,1.2) and can also be
verified by comparing coefficients of like powers of x. The formulas are
as follows:
(2) (¢- a) O(@-) + (2a~c +x) 6~a G(a4)=0,
(3) ele -Y) O(c -)- cle -1+x)O+ (c ~ a) x O(c +) =0,
(4) (a-c+)) @-a (a+) +(c - 1) O(c -) =0,
(5) ¢®-c @(a-)-x O(c +)=0,
(6) c(a+x) O-(c—a) x @(c +) ~ae G{a +) = 0,
(7) (a- 142) 6+ (c ~a) O(a-)-(c-) O(c -) = 0.
These relations are not all independent. From two suitably chosen ones,
e.g., from (2) and (4), all the others follow by simple operations.
‘Any function O(a +m, c +n3x),m, n, integers, is said to be asso-
ciated with (a, c; x). By repeated application of the relations between
contiguous functions, it is easy to prove that any three associated func-
tions are connected by a homogeneous linear relation whose coefficients
are polynomials in x.
By means of term-by-term differentiation, we have
d a
8) —O(, ¢5 )=— O@+lc+h 2)
dx e
or @' = (a/c) O(a 4, c +), By repeated application, the derivative of any
order of © is an associated function, and hence any three derivatives
are connected by a homogeneous linear relation whose coefficients are
polynomials in x, The differential equation 6,1(2) is the simplest exe
ample of such a relation. Combining (8) with the relations between con-
tiguous functions, we have
[© - ®(c-1)]
(9) oa (oa) -e1-(E- ) O(c N4O=
: c
Further useful formulas are:
ap
dx”
(10)
(a),
O(a, c;x)=—* B(a+n, c +n; x)
©,
n
(ul) = [x2 O(a, oj I = (a), 27? O(a tn, o; x)
fe65 CONFLUENT HYPERGEOMETRIC FUNCTIONS 255
a
02)
[xe7! &(q, ¢; x)] =(-I)"(1—e), x67!" O(a, c —n; x)
(c-a),
a3) = [e~* ®(a, ¢; x)] =(-1)" e7* O(a, c +n; x)
i) = [en# 2-21 Ba, 05 MI =(e =a), em xe! G(a=a, 65 2).
7"
Tla+n)
Tla+n
(= (= ala+D ++ (a4n- 1) =
n=1, 2,3, 0008
Basic integral representations
It is known that homogeneous linear differential equations whose
coefficients are linear functions of the independent variable can be in-
tegrated by Laplace integrals. The integral representation
The) ph yet yyene! dy
(1) ®(@, ¢; x) = ————___ °
T(@) Pe -a) Rec>Rea>0
can be verified at once by expanding e* in powers of x. This represen-
tation suggests the integration of 6,1 (2) by an integral of the form
ya Sener (4 eet de,
which is also suggested by Laplace’s method. Substituting this in 6,1 (2)
we use the identity
ae d wat pon ena~
E a) a | [ieee toe) (+ ences]
aden)
to show that the integral satisfies 6.1(2) provided that C does not pass
through any singular point of the integrand, and that the initial and final
values of e7*¢ (1+ 1)°~® are finite and equal to one another. If Re ¢>
Re a> 0, one possible choice for C is the interval (0,-1), thus showing
that (1) satisfies 6.1(2). If Re a> 0, another possible choice is a ray
from the origin. We put
1 co
jo watyont evant i
-(2) ¥G, ¢; x) re de eee" (142) dt Rea>0.
This equation defines a solution of 6.1(2) in the half-plane Re x > 0.256 SPECIAL FUNCTIONS 6.5
‘The domain of definition can be extended by rotating the path of inte
gration. Thus,
@) VG ce [PO mttaer ae
eT) “0
Rea>0, -r 0. This proves (10)
Lwith |p,,,| instead of O(|e|~*"') for a sufficiently large value of n],
But then (10) is clearly also valid forn = 1, 2, ..., since each term of
the asymptotic series in (10) behaves like |p,,,|. For more general
results see T. M. MacRobert (1923) who has proved (10) for a range of
arg ¢ which is > 7.
If a, ¢ and z are fixed numbers, c #0, -1,-2,... and 0 <|2| < 1, and
if |b| + © such that ~37/2 < arg zb < %n, then we have
ce bz)"
(13) F(a, 6; ¢; 2) = F(a,b; ¢; vy-| 2 or | [1+0( 8/791
ae
n=o
and here the asymptotic fomulas for the confluent hypergeometric func
tion of a large argument (cf. Chapter 6 or Whittaker-Watson 1927, 16.3)
give
(14) F (a,b; €5 2) =e ™9[P()/T(c — )].{52)~ 2 [1 + O([b2|-"))
+[P()/P(a)] e® (b2) [1 + O(|bz\7")].
Similarly, if - 4m < arg bz < 37/2, ne have
(15) F (a,b; ¢; 2) =e!" *{ 1 (e)/E (e — a] (b2)~*[1 + O(|b2|~1)]
+[I(O/T(a) e* (52) [1 + OC \bz|")1
In the case where more than one of the parameters tends to infinity,
G. N. Watson (19 18) has obtained the following results.
Let ébe defined by z +(z?- )* =e #8, and put
1-ef =(ef =) e*
where the apper or lower sign is taken according as Im z % 0. Then for
large | 4{ we have
(16) (42-4 TA Flat dy a-c+ 14+ As a-b4+ 142d; 21-271]
_ 2 Pa-b + 1420100 A elt NE
T@-e+1+P(e-b+a)
x (L= ef )-eth (14 oF) eve FHK [1 40( 479]
where |arg 4} < 7-5, 8>0 and also
(17) Fla+a, 6- A; e3%-%2)
Ree) a
a Gee
Toles
x NHL AA OE 4 eb imteK) oAM EIT. OIA")78 SPECIAL FUNCTIONS 2322
where the upper or lower sign is taken according as Im z 2 0 and where
|Al is large, €= C+ in,
~ Wan w,+ 8 1, In certain cases,
quadratic and cubic transformations exist for ,F,, e-8.,
EF @, b, cr 2
Ms" tyya-blsane
“0-37 F, a ae
(Whipple 1927) and
® GF, [* 2b-a-1, 2-244; |
l+a-b, l+an-c
bab + 9/2
a a/3, a/3 + 1/3, a/3 + 2/3; ~27 z/[4(1-z)*]
ee b,a-6+3/2
which have been proved by Bailey (1929). Bailey also gave a linear
transformation of a nearly~poised ,F, of the first kind:
Qa-La+%a-b-%y2
@ G-pe 44%, :
2 a a-%a+b+% ]
7 26-1, b+%,b~a-%s
=(1-2)27) .
=(1-2) Al b-% a+b 4%
Burchnall (1949) has proved that a well-poised ,F, of argument x can
be expressed as a sum of series ,F, of argument ~ (x ~ 1)7/(4),
‘There are also a few cases known where a ,,,F, can be evaluated
for an argument # I, e.g.,
() OF a, 1+%a, b, c3-1 T(l+e-a)P(l+a~c)
43 | Ma, l+a-b, l+a~c TO +a)C0+a-b~c)
(cf, Bailey 1935, p. 28, and also Bailey 1929).
Some special cases: The formulas developed in section 4.4 and in the
present section contain alarge number of results which cannot easily be
proved directly and which have found much attention in the literature,45 GENERALIZED HYPERGEOMETRIC SERIES 191
For instance, 4+4(5) gives (with a=b =e =-n):
LT ya\ |? _ cos Kam P04) Th + Yn) PC + 8n/2) 2"
® fen [0] ~~ [Pd +ka) Pasha al
If-we substitute b = 1+a+7 in 4,4(8) and let x + 0 we obtain
a, 1+ ac dye
© sf Ee l+a-c, l+a-d, eee]
Tt+a-c) TI +a-d TA +a-e)P(l+a-c
“PO +aP+a-d-e)P+a-c-e)P+a
A special case of this is the Dougall-Ramanujan formula
2,69), 62),
G+s),+),a+2),
() 142 2
_PG@+ DTG + DPG+DT@+ytz2+D
—Ty+zt+DP@+z2+ DP @ty+D
which is valid for Re(x + y + z + 1) > 0, Another consequence of (6)
Gibaslesl 27-2 1)1
z-l_, &-De-9
3 ee
2 Cer E)
‘The series on the left-hand side converges if Re x > 1. A large number
of other special cases is given by Bailey (1935, Examples pp. 96, 97) and
Hardy (1923).
The truncated hypergeometric series
(@), 6)
Gee). eee
a ss
soe,
can be expressed in two different ways in terms of oF et
Spe be| | =n a be
and
ae @), no z [* nn, l-n-c3 a
| (emt l-n-a,1-n-5
If z = 1 we have
yq(@ 6 ol =
Ti+n+)DPG+n+) a, b,c+n;
niT(a+b+ne a"2]catbtn+l]192 SPECIAL FUNCTIONS 45
_TQ+a-c) PA +b-e)
“TQ=e)TG-c+a+d)
xd p- nt Ong op [Poa l-b aes
(@+DIC~D FAL 2cn+2
Tatn+ DEG +24) p [ore ene ta
| eee ee
alP@a+b-c+Dlernal 2
(cf. Bailey 1935, pp. 93, 94). Corresponding formulas exist for the trun-
cated ,F, and a special truncated ,F, with unit argument (cf. Bailey
1935, pp. 94, 95).
4.6. Integrals
The analogue of the Eulerian integral representations in section
2413, in the case of the general ,F, has been given by Pochhammer
(1893b), Erdelyi (1937), Jef. also’ 5.2(2)] and by Pevnyi (1940) for
integral relations of the type given in 24(2). The extension of Bames’
integral representation 2.1(15) to the case of a general ,F, cam be
derived from the results in sections 5.3 and 5,6.
Many definite integrals can be expressed by one or several series
F, . In particular, the Laplace transform of a ,F,, p< 9, is
3 gly eae
rast ato | MO a,3* |
Py Pyreees Pg
A formula by Shanker (1946) gives the Hankel-transform of a ,F,:
Se PF Lp, Kp + m4 Ds Hlp—m 4D, Hlp+y4 Ds ye]
cc+n+]
sie
x d,(2t) (2t)* de =(-1)" 2°
x ,F, [p, 4p + m+); lp m+ Ds Kp tv + D5 -%27 1,
where
m=0,1,2,+4., and Re(p +y+1)>0, Re(p—m+1)>0, Re p> %
For other special results see section 4.7 and Erdélyi (1938).
4.7, Various special results
H. Bateman (1933, 1934) and Pastemack (1939) investigated the
polynomials in z,
Fz) = ,F, Cra+) 4+423 1; D n=0,1, 2, e085AT GENERALIZED HYPERGEOMETRIC SERIES 193
and Bateman (1936) studied the polynomials Z, (z) and J” (z) defined
by Yol
erate.
Z,(2)=,F,emntl lite) a
Tne leu)
nlP@+DP@+14 hu)!
Bateman’s results have been generalized by Pasternack (1937, 1939)
and by S. 0. Rice (1939), The latter introduced
HC, pv) = Fy (+n, n+ 1, 3 1, ps v)
278 Jte¥(2) = Fylonzut vt Le %us 2%),
where n= 0, 1,2, «+ and &, p, », are complex variables but p #—n ~ 1,
-n—-2,.... Rice found that
H, Ag; p. »)
Tp) i
=< €-¥(1—2)P-E“* P (1 Qv0) di
atte f L a) a wt) dt
Rep>Re €>0, P,(z)=,F, (
lifp=q+1L
For p> q + 1 we have
(2) Elpsa,: 93 p,:x)=
i Ta,-a)
=) o——_ Pra)
a
=) Tt Pp,-a)
a204 SPECIAL FUNCTIONS
VF yaplays a, py ty ees ,- pth;
Om Oy errr ar eeey ama, th CDP x]
where |x| < Lif p= +1. The prime in I’ indicates the omission of the
factar (a, — a,); the asterisk in F the omission of the parameter a,~ a, +
1; an empty product is to be interpreted as 1; and zero or negative in-
teger values of the a, are tacitly excluded. Similar conventions will be
observed throughout this chapter. The asymptotic expansion, for x + 2,
- Ap - q+) aK ag x < 4(p - ¢ + 1) a, of (2) is given by the right-
hand side of (1). Originally (MacRobert, 1937-1938),£ was defined by
the multiple integral
Ta,
3) Eps aj: 93 p,: x)=
Gree a)
4
xu f, NO ad) Hdd,
cae te
promt
x
nA, O47)
‘aty \ “ty
re Ady dy at
vee
oe A eas
Bye 1 eater dd
cL es [ * Ta) Otay Fa, °
where arg x| <1 p>q+ land the a, and p, are such that the integrals
are convergent. The equivalence of (2) and (3) can be proved (Mac
Robert 1937-1938).
Other related functions are the P-function which arises when (2) is
written as
P
(4) E(psa,:93p,:2)= x, Plas p-a,: 95 py: %)
and two further functions, denoted by Q and H, which are multiples of
P and E ae, (MacRobert 1937-1938). An alternative notation
for E(p3 ay: q3 pi %) is E (ays vey Oy! Pys sees Pg 2
Several beecie cases of the E-function arise from (1) by means of
the formulas developed or quoted in Chapter 4. From (2) and (3) some
further interesting particular cases arise among which the most important
are the expression of the modified Bessel function of the third kind,
(5) (na) K(x) =e7* cos(vn) E(% +4 4 ~ v2 2x)5.2.2. GENERALIZATIONS OF THE HYPERGEOMETRIC FUNCTION 205
{MacRobert 1937-1938 (12)], and that of Whittaker’s W-function
(6) TK-k-mTK~kam) xt eM, | (x)
=EQ-k-m, 4-k+m: ix)
[ MacRobert 1941(25)]. It may be noted that V, , (ix) W, , (ix) may also
be expressed in tems of the E-function [ MacRohert 1941 (15')] as can
other combinations. The £-function itself is a particular case of Meijer’s
G-function [cf. 5.6(2))-
5.2.1. Recurrence relations
A basic system of recurrence formulas has been given by MacRobert
[ 1941, equations (20) to (24)]. The three most important formulas are
CD) ay x Elay seers Gps pyr eves Pgt ®)
=e Ela, ty ayy sees Opt Pps eve y Py tA)
+Elay+Laptleesa,thpy thes pet hay
d
8) Ely ees at Pry ene 9 Pgt AD
=x? Ela, +l eee, a,th ppt lees pth)
and
(9) (py Vx E lay, e+ a, pyre
+ qi)
= Ela,,...,0,:p Veer sees Ogi x)
ee ee
5.2.2, Integrals
A few integrals with hypergeometric functions may be evaluated in
terms of the E-function. A typical example [ MacRobert 1937-1938 (21)]
is
ao) * eth) LF, (ay B; 8; -d) dr
T() 277
r@ )
which may be proved by substituting
oF, (a, Bs 8s ~X)
e ae
a So uPA (Ley? sar] dy
E(a, B, y: 8: x) Rex>0, Rey>0,
T'(6) '(8- p)206 SPECIAL FUNCTIONS 5.2.2
on the left-hand ‘side and using (3).
The integral
Sf e®® Fla, Bi ns OWE
may be evaluated by complex integration; we integrate around a contour
which is a rectangle with vertices at 0, 1, 1+ ik, ik(k > 0 )and which
has indentations at 0 and 1 and then make the radii of the indentations
tend to zero and k » «, using (10). [MacRobert 1937-1938 (23); there are
also a few other integrals in this paper. ]
Integrals with the E-function are discussed in Mac Robert 1941. The
most important are the integrals of the Laplace or Euler type, but there
are some more general integrals too. Typical samples are
1 tot)
oa OF EE ays vee yt pay ee gt Ea) dE
—mgagesa,
MEG, y o00y Opt Pyy ves Pg 2 *)
(12) JP eA AP ECG, cee yt Dyr eee Dgt A/A) GA
Ely ose yyy Bt pps 000 9 Pgt®) Re B>0,
(ot)
(18) Jo e® Et Blayy vee 5 yt ys vee Pgs 3/E) dE
=e¥"E(a + ys Bt Pps o0+ y Pg 28%)
we OTE (ayy sees ys Bt Pas tease).
If p=q+1, this holds provided that ¢= x inside the loop; ifp
1, the right-hand side reduces to
24 sin(Bm) Bays 01+ 4 5) Bi Py eee Pgi 2
(14) SP ATE LAN EL ayy soe yay Pyy eee Pg? +2]
=P (o~B) Ela, 0+ 5 ayy Bt yy vee» Pgy 0 *)
see errta! Re(B)>0, Re (a-f)>0.
The proof of these formulas is based on the definition of the E-
function; (1) or (2) and (3) are used according to the values of p and g.
MEIJER’S G-FUNCTION
5.3. Definition of the G -function
Meijer’s G-function provides an interpretation of the symbol ,F,5.3 GENERALIZATIONS OF THE HYPERGEOMETRIC FUNCTION 207
when p > q + 1; this interpretation is in complete agreement with the one
given by MacRobert’s E-function. In addition, all significant particular
solutions of a hypergeometric differential equation 5.1(6) may be ex-
pressed in terms of the G-function.
Originally (Meijer 1936), the G-function was defined in a manner
resembling 5.2(2). Later (Meijer 1941c, 1946), this definition was re-
placed by one in terms of Mellin-Bames type integrals (for which see
section 1.19). The latter definition has the advantage that it allows a
greater freedom in the relative values of p and g. Here we shall complete
Meijer’s definition so as to include all values of p and q without placing
any (non-trivial) restriction on m and n.
We define
o ese (fet)
A, T;-s) A, T(1-a;+s)
ads
ma is
i ra- te a, (Leas)
j=att J
where an empty product is het as 1,0
lLaendeitherp > 9 orp=q and |x} >L208 SPECIAL FUNCTIONS 5.3
We shall always assume that the values of the parameters and of the
variable x are such that at least one of the three definitions (2), (3), (4)
makes sense. In cases when more than one of these definitions make
sense, they lead to the same result so that there will be no ambiguity
involved.
The G-function is an analytic function of x; it is symmetric in the
parameters @. 1 @,, likewise in @,4,1 +++) py in by, «++,» and
inb,
With (3) the idtegral can be evaluated as a sum of residues. If no two
, j=1,+++>m, differby an integer, all poles are of the first order and
= Ct
A re,~s) Brat e-ap
P
i Pp
_i ] oT se T'(a,-6,)
x Fy, [1+ b,- 5
Lib, m dys eceg acces L4b, = 55 (DPA a]
pP
L
y14b)-a,3
sy lta,-a,3 CTs
In these expansions conventions corresponding to those of 5,2 (2)
hold.5.3.1 GENERALIZATIONS OF THE HYPERGEOMETRIC FUNCTION 209
5.3.1. Simple identities
If one of the a, j= 1, ... , 7, is equal to one of the b., j=m+1,
visa q(orone of the b,, j= 1,...,m, equals one of the a, j=an+ 1,
s+» P)y then the G-fimction reduces to one of lower order: p, g and
n (and m) decrease by unity.
mex, cet ee (eferrts)
™ prq>)
is such areductionformula, and all others are similar.
Chvious changesof the variable in the integral (1) give
o cant |r \ gan re
2) cal: s) Om ( re).
nm co To.
(9) om =cm **)
r
(10) on(- 2 see evrote!
= (Qn)Kethar ann gkp-katimayn 884 a, t,t tb,
p24 4b, 4b, +%
In (10) the duplication formula of the gamma function has been used,
and there is a corresponding formula which uses the multiplication
theorem 1,2(11) of the gamma function.
The most important of these formulas is (9) because it enables as to
transform a G-function with p > to one with p
1,
@,\_1 Tibaey om ttn =m [97
| msq-l
P a
() (neers Th S-a4D- Mh (9-o)]y=0 Sax,
jst a pat i dx
Clearly, every differential equation of the form 5,2(6) may be reduced
to this form by a change of variable. Equation (1) is of degree max (p, q)
and on account of 5,3(9) we may assume p < q. The solutions of (1)
have been investigated by Meijer (1946).
If p
p +2. Then, if is in the sector determined by (3) and (4),
the p functions
(5) Ge (xterra yr |The Fae rere Bpagt Fatty vee a, )
b b
1g
hades es
and the g — p functions
(6) Gre (- Geomcn- ahd
form a fundamental system of solutions of (1),
If p = q, the point x = © is also a singularity of the regular type
and, with the conditions (3), a fundamental system is given by (5). In
this case x = (-1)?""~" is also a regular singularity, but no funda-
mental system for the neighborhood of this point has been given in the
literature.
Ifp > q,5.3(9) may be used in order to reduce the differential equa-
tion to the case already discussed; x = 0 and x = o interchange their
a
") h=g, gt+h-.-,g+q-p+l
role.
In any case, it is clear from (1) that for fixed p, q ay) --. 5 Gp»
Dyy sees bay all @ +1) (q+ D functions
@. CoC hee] Osmsq Oxnsp
satisfy the same’ differential equation.
5.4.1. Asymptotic expansions
We shall assume p
¢ can be obtained from
5.3(9) by interchanging the role of x = 0 and x = o. Certain integer
values of combinations of parameters are excluded, and so are a few212 SPECIAL FUNCTIONS 5.4.1
other exceptional cases, for example, those in which the coefficient
of the dominant term of the asymptotic expansion is zero.
The point x = 0 is a singular point of the differential equation (1), and
the behavior of G (x)in the neighborhood of this point follows from 5.3(5)-
We have
(8) Grr (x)=O(k|4) as x90
\
where p
1 m+n>kp+%q, and larg x|<(m+n—%p—'q) am;
or if
(10) g=p+1, & is some integer, and
largx—(m+n—p+2k—-Vn|<¥n
As x > 0, p
¥p+khq n=0, jargx|<(m-%p-~q)m
When +0 and p
%p+%q and fargx|>(m+n-%p-\q) a
or
(ii) m+n ¢ %p+%q and no restriction on arg x.
(13) Ifq =p +1, letk be an integer, and let
Jarg x —(m-+n—p + 2k)n| < hm
In this case we must have either
(Gi) m+n2p+1 and k20 of kg p-m-n
or
(ii) m+n p and no restriction on the integer &
When p =, the behavior of G(x) as x -» & follows from 5,3(6); it is
that of a power of x.
For more complete and more detailed results see Meijer 1946 $18.5.5.1 GENERALIZATIONS OF THE HYPERGEOMETRIC FUNCTION 213
5.5. Series and integrals
Of the more involved functional relations for the G-function the mast
important ones are series and integrals, Comparatively few series of
G-functions have been investigated. The number of known integrals
is, however, very considerable. Only a féw samples will be given here;
for more details refer to the original papers, mostly by C. S. Meijer.
5.5.1. Series of G -functions
A first group of series expansions in terms of G -functions consists
of the four multiplication theorems (Meijer 1941 c),
cer( ae [Or oe
Pa Byseees by
b - Gye @
poo bo ee vere,
o Aan” Ce)
A-1U